searches for gravitational wave bursts : methods and challenges

G080404-00-Z 1MIT Seminar

Searches forGravitational Wave Bursts :

Methods and Challenges Shantanu Desai

Pennsylvania State University

2

Outline

• Basics of Signal Detection as applied to LIGO • Similarity with other fields• “Nuts and Bolts” of LIGO Burst Searches• Examples of S5 Transient Noise Events • Aperture synthesis techniques (brief)• Connection with Dark Matter conundrum

This talk will not present any results from burst searches, but only the methods used.

3

Jargon used in this talk

• ETG Event Trigger Generator (algorithm for burst searches)

• IFO Interferometer• H1 4 km LIGO Hanford detector• H2 2 km LIGO Hanford detector• L1 4 km LIGO Livingston detector• Glitch Noise transient

4

• Compact Binary Inspirals :Template based searches for merger of neutron star/black hole based binaries

• Unmodelled burst sources : Short duration transients(< 1sec) without any knowledge of waveform (core-collapse SN, GRBs etc)

• Periodic sources :Known and unknown pulsars in our galaxy

• Stochastic Background :Search for cosmological background from a variety of early universe processes.

LIGO Science Analysis Efforts

NASA WMAP

SN1987A

5

Basics of Signal Detection

A. Estimate number of signal events B. Calculate the expected background

Based on (A) and (B), make a decision on whether there is a detection or if expected signal is consistent with backgroundAnd there is no detection.

In case of a claimed detection you need to asses the ``statistical significance’’

IMPORTANT: Decision on detection/no-detection is not a binarystatement . Explicit calculation of the probability of the signal being due to fluctuation of the background must be evaluated.

6

Similarity to other fieldsIssues in LIGO data analysis techniques similar to

• TeV -ray astronomy• Neutrino astronomy

and High-energy physics experiments such as searches for • Proton decay • Magnetic monopoles (1982 Cabrera event)• Higgs boson (2000 LEP results)• Neutrino-less double-beta decay• Dark matter detection• Super-symmetry• Fractionally charged particles, etc

No detection sofar (somewrong claimsof detection)

Important to keep track of how the data analysisin the above searches are done.

7

Philosophy in LIGO Data Analysis

Surely you are joking Mr. Feynman (1985)

Blind Analysis done in LIGO (unlike in astronomy) usually doneIn high energy physics experiments.Look at the observed signal events ONLY after thebackground has been tuned and fixed.

8

Incorrect claims of detection

Above claim not generally accepted as it is ruled out by other sensitive dark matter experiments even though the claimed significance is ~ 8

Dark Matter: R. Bernabei et al 0804.2741tro-ph/0307403

9

Incorrect claims of “excess”

Gravitational Waves

See L.S. Finn : gr-qc/0301092 for a critique of the above result

Important to understand sources of background and Do various cross-checks in case of something interesting.

P. Astone et al gr-qc/0301092

10

• All-sky all-waveform searches (“untriggered searches”) at all times

• Triggered Searches : Look for gravitational wave signals associated with electromagnetic triggers

Classification of Burst Searches

11

Rudiments of LIGO Data

• Data from auxiliary control channels in the detector and various environmental monitors (eg. seismometers, magnetometers) also stored in similar “channels’’ for diagnostic and off-line trouble-shooting.

• LIGO data containing possible gravitational wave signal is sampled at 16 KHz and digitized in a data acquisition unit called gravitational wave channel

12

Conditioning of Data• Raw LIGO data needs to be conditioned and this involves reducing the sampling rate, removal of lines from PSD and whitening the data (make the PSD flat)

T. Summerscales thesis (2006) S3 data

• Data conditioning implemented in different ways by various groups.

13

Basics of Burst Searches

• Various algorithms (time-domain, wavelets , etc) are used to look for transients (“triggers”) in the gravity wave channel.

TriggerDuration ~ 0.2 s

Wh

itene

d T

ime

-ser

ies

Time (seconds)

• Same algorithms can also be applied to look for transients in auxiliary channels.

14

Algorithms used in Burst Searches

LSC Burst group has developed multiple (> 4) algorithms with the SAME

science goals :

Different burst algorithms see different events.

A. Stuver Ph.D. thesis (2006)

• WaveBurst• Block-Normal (based on Bayesian change point algorithm)• Kleine Welle• Q pipeline • Excess Power• Slope• TFClusters• Hilbert-Huang Transform and many many more

Injected signal

Eventthreshold

15

Flowchart of Burst Analysis

Generate Triggers (for all 3 detectors)

Eliminate/veto triggers due to transient noise

Coincidence and Consistency test

+/- 30 ms time-window Many independent time-lags

A : signal B : background

Compare (A) to (B) and you are done. Efficiency estimated byinjecting various ad-hoc signals in the pipelines.

16

Transient Noise Identification

Various methods have been used for studying the cause of transient sources of noise lasting few milli-seconds (called “glitches’”)

• Event Visualization tools

• Statistical methods (MIT/Syracuse/UMD)

• Measured Transfer Functions (R. Schofield, environmental)

• Expertise of Commissioners

• Listening to Glitches (Syracuse)

• Multi-dimensional classification of noise triggers (UTB)

For more details on methods used see Blackburn et al : 0804.0800 Gouaty et al : 0805.2412

17

Event Visualization tool (I)

Filtered Time-Series + Median normalized spectrogram

QuickTime™ and aTIFF (Uncompressed) decompressor

are needed to see this picture.

Glitch No Glitch

Time (seconds)

K. Rawlins (2005)

Fre

quen

cy (

Hz)

18

Event Visualization tool (II)

Q-Scan Look at the projected detector data with Q-transform as basis.

QuickTime™ and aTIFF (Uncompressed) decompressor

are needed to see this picture.

QuickTime™ and aTIFF (Uncompressed) decompressor

are needed to see this picture.

Glitch No Glitch

This tool also used by operators, detector experts and in GEO and VIRGO

S. Chatterji Ph.D thesis (2005)

19

Noise Transient : Seismic Noise

Excess Seismic noise

Hanford Y-end seismometer

Time [seconds]

Fre

quen

cy [H

z]

Gravity Wave Channel

• Transient seismic noise < 10 Hz getting up-converted to ~ 100 Hz in the gravitational wave channel.

20

QuickTime™ and aTIFF (Uncompressed) decompressor

are needed to see this picture.

Fre

quen

cy [H

z]Noise Transient : Acoustic Noise

Time [seconds]

Livingston Y-end microphone

Time [seconds]

Gravity Wave channel

Multi-tone feature most probably caused by an overflying helicopter

Multi-tone features in the microphones

21

Noise Transient: Power spikes

• Disturbances on power mains These cause simultaneous coherent noise transients in both LHO 4km and 2 km detector

Mains power voltagemonitor

Gravity Wave channel

H1 H2

22

Noise Transient: TCS

• TCS glitches caused by mode hops in the TCS laser in which there is a sharp drop in power level incident on the interferometers. Seen in all H1, H2 and L1 during S5. More problematic at LHO

ThermalCompensationSystem

Gravitywavechannel

23

Noise Transient: ISS

• Causes very loud transients in gravitational wave channel.• Only seen in H1 during S5.

24

Noise Transient: OSEM glitches

• Seen in H2 (1st 3-4 months). Fixed by S. Waldman

Gravity wave channel

OSEM channel

25

Noise Transient: Optical Lever

• Such glitches happen when optical lever lasers need to be replaced.

Gravity wave channelBS optical lever sum

26

Gravity wave channel

Noise Transient: Calibration Line Glitch

Dropouts in injectedCalibration signal

27

Noise Transient: X-end processor failure

DC light levelIn X-arm

DC light levelIn Y-arm

Gravity wavechannel

28

Noise Transient: Computer Malfunction

L1 triggers Peak when time (in minutes) ~ 0

• Hourly noise transients first appeared on October 3rd 2006.

• Attributed to snapshot processes performed by the detector DAQ on a periodic basis (every hour in Oct. 2006) called autoburt.

29

Noise Transient: Tidal Desaturation

Events in Gravity wavechannel

Enhanced glitchiness

• Interferometer becomes glitchy when the data from the tidal servo comes out/goes into maximum absolute value of 90 counts. Effect mainly seen in H1 during S5

30

Noise Transient: Data Acquisition Problem

DARM-ERR AS-Q

• Probably caused by a timing problem when DAQ is unable to keep up with the data-stream. • Seen in all 3 interferometers (very low dead-time)

31

Noise Transient: Dust

• Signals from PD 1 and 4 inverted compared to PD2 and 3

• Possible cause of such glitches is due to dust along the beam path and verified through dust injections (R. Schofield)

• These glitches not seen in any auxiliary channels. Monitor written by for such events. (J. Zweizig)

Event found by Black Hole Ring-down search

PD1 invertedw.r.t PD2

32

Noise Transients Recap

• Source of many noise transient events unknown. Lots of work still in progress in hunting down all noise transients in S5.

An example of anunknown H2glitch in June 2007

• A database of noisy intervals maintained by K. Riles : http://www-mhp.physics.lsa.umich.edu/~keithr/S5DQ/flaginfo.html

33

Use of Vetoes in Analysis

• Generate Data quality flags for bad intervals with different severity levels.

Category 1 - Do not analyze Category 2 - Used in post-processing Category 3 - Advisory for detection confidence and used in upper limit, if no detection Category 4 - Advisory flag used to exert caution in case of a detection candidate

• Use vetoes from auxiliary channel on an event-by-event basisGravity wave channel

Potential veto channel

• Check a real gravitational wave would not couple to veto channels.

34

Aperture synthesis methods

Network analysis combine data coherently from various detectors

However antenna response matric for near-aligned detectors is ill-conditioned (Condition number >> 1)

(2 polarizations)

Ω h2

See papers by Ajith, Chatterji, Finn, Hayama, Klimenko, Lazzarini, Mohanty, Rakhmanov, Schutz, Searle, Stein, Summerscales, Sutton, Wen, and many more

35

Effect of Regulator

Numerical simulation for LIGO + VIRGO 1000 trials 2 polarization waveforms

True RA = 40°, DEC = 98°

Without regulator with (Tikhonov) regulator

36

Skymaps with Simulated Signals and Noise

QuickTime™ and aTIFF (Uncompressed) decompressor

are needed to see this picture.

Averaged noise Skymap for a LIGOonly network

QuickTime™ and aTIFF (Uncompressed) decompressor

are needed to see this picture.

Skymap at location ofsignal for a LIGO onlynetwork

37

Possible glitch in gravity in ultra-weak field ?

• General relativity agrees very well with observations at solar system and binary pulsar length scales.

• At longer length scales, 95 % of the universe is made up of two ``dark’’ components.

• Evidence for dark matter comes from galactic rotation curves, gravitational lensing, large scale structure, etc. No clue on its identity.

• Could dark matter be a consequence of modified gravity (ala Vulcan)?

38

Gamma-Ray Bursts Soft gamma ray repeaters Pulsar glitches Low-mass x-ray binaries Neutrino triggers Optical transients Core - collapse supernova Blazar flares

Search is done by looking for gravitational waves in a narrow time window around the trigger (~ 100 seconds)

LHO

LLO

Swift/HETE-2/

IPN/INTEGRAL

RXTE/RHESSI

arXiv : 0802.4320 (Abbott et al)

Triggered Gravitational wave searches

The ansatz assumes that propagation time for GWs is same as light.

One pre-requisite for this assumption is that Dark Matter exists

Cr: Z.Marka

39

Time delay of light due to its passage near a massive body first

calculated by I. Shapiro (1964). This delay is ubiquitous (radar ranging, binary pulsars etc) Light Ray Mass

Gravitational waves and neutrinos also experience same Shapiro delay as light in general relativity.

Shapiro Delay

Total (GW/photon/neutrino) travel time for explosive events =

light travel time + Shapiro delay from intervening mass.

Shapiro delay for SN1987A (50 kpc) ~ 5 months (M. Longo 1987)

Mass of our galaxy is dominated by dark matter which is dominantcontribution to Shapiro delay for any nearby transient sources

40

Modification to Newtonian Gravity

• Need for dark matter arises only for ao < 10-8 cm /s2

• Tully - Fisher relation : L V4

a = anewt (anewt/a0)-1/2 for a < a0

astro-ph/0507589astro-ph/0204521

Slope = 3.9 ± 0.2

Milgrom (1983)

41

1957 Zwicky

1983 Milgrom (Modified gravity to explain rotation curves & Tully-Fisher relation)

1984 Milgrom and Bekenstein (Non-relativistic generalization)

2003 Soussa and Woodard (No-go theorem)

2004 Bekenstein (Relativistic theory of MOND : TeVeS)

2005 Moffat (Another Relativistic theory to avoid dark matter)

2006 Skordis et al. (Cosmology of TeVeS not so bad)

• Kahya and Woodard (Model-independent test with gravitational waves)

History of modifications to GR for DM

42

Model-Independent Test

For a whole class of modified gravity models which avoid dark matter :

Shapiro Delay for light/neutrinos = Potential of visible + dark matter.

Shapiro Delay for gravity waves = Potential of visible matter only.

Gravitational waves will earlier compared to light. Time delay (in days) for 3 sources below

Source Distance NFW Isothermal Moore

Sco-X1 2.8 kpc 4.88 4.98 4.97

SN1987a 51 kpc 74.8 78.2 74.5

GRB 070201 780 kpc 804 742 811

arXiv: 0804.3804

Simultaneous detection of gravitational waves and photonswill rule out these models.

43

Conclusions

• Multiple methods developed for burst searches to look for transients as well as to do coherent network analysis.

• Many new transient sources of noise seen during S5. However cause of many glitches still unknown.

• Gravitational wave observations could resolve the 75 year old dark matter conundrum.

Acknowledgments :S. Finn, Keith Thorne, A. Stuver, T. SummerscalesLSC Glitch Group (~35 active members) K. Hayama, S. Mohanty, M. RakhmanovE. Kahya, R. Woodard